arch/powerpc/platforms/pseries/lpar.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * pSeries_lpar.c
   4  * Copyright (C) 2001 Todd Inglett, IBM Corporation
   5  *
   6  * pSeries LPAR support.
   7  */
   8
   9 /* Enables debugging of low-level hash table routines - careful! */
  10 #undef DEBUG
  11 #define pr_fmt(fmt) "lpar: " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/dma-mapping.h>
  15 #include <linux/console.h>
  16 #include <linux/export.h>
  17 #include <linux/jump_label.h>
  18 #include <linux/delay.h>
  19 #include <linux/stop_machine.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/cpuhotplug.h>
  22 #include <linux/workqueue.h>
  23 #include <linux/proc_fs.h>
  24 #include <linux/pgtable.h>
  25 #include <linux/debugfs.h>
  26
  27 #include <asm/processor.h>
  28 #include <asm/mmu.h>
  29 #include <asm/page.h>
  30 #include <asm/setup.h>
  31 #include <asm/mmu_context.h>
  32 #include <asm/iommu.h>
  33 #include <asm/tlb.h>
  34 #include <asm/cputable.h>
  35 #include <asm/papr-sysparm.h>
  36 #include <asm/udbg.h>
  37 #include <asm/smp.h>
  38 #include <asm/trace.h>
  39 #include <asm/firmware.h>
  40 #include <asm/plpar_wrappers.h>
  41 #include <asm/kexec.h>
  42 #include <asm/fadump.h>
  43 #include <asm/dtl.h>
  44 #include <asm/vphn.h>
  45
  46 #include "pseries.h"
  47
  48 /* Flag bits for H_BULK_REMOVE */
  49 #define HBR_REQUEST     0x4000000000000000UL
  50 #define HBR_RESPONSE    0x8000000000000000UL
  51 #define HBR_END         0xc000000000000000UL
  52 #define HBR_AVPN        0x0200000000000000UL
  53 #define HBR_ANDCOND     0x0100000000000000UL
  54
  55
  56 /* in hvCall.S */
  57 EXPORT_SYMBOL(plpar_hcall);
  58 EXPORT_SYMBOL(plpar_hcall9);
  59 EXPORT_SYMBOL(plpar_hcall_norets);
  60
  61 #ifdef CONFIG_PPC_64S_HASH_MMU
  62 /*
  63  * H_BLOCK_REMOVE supported block size for this page size in segment who's base
  64  * page size is that page size.
  65  *
  66  * The first index is the segment base page size, the second one is the actual
  67  * page size.
  68  */
  69 static int hblkrm_size[MMU_PAGE_COUNT][MMU_PAGE_COUNT] __ro_after_init;
  70 #endif
  71
  72 /*
  73  * Due to the involved complexity, and that the current hypervisor is only
  74  * returning this value or 0, we are limiting the support of the H_BLOCK_REMOVE
  75  * buffer size to 8 size block.
  76  */
  77 #define HBLKRM_SUPPORTED_BLOCK_SIZE 8
  78
  79 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
  80 static u8 dtl_mask = DTL_LOG_PREEMPT;
  81 #else
  82 static u8 dtl_mask;
  83 #endif
  84
  85 void alloc_dtl_buffers(unsigned long *time_limit)
  86 {
  87         int cpu;
  88         struct paca_struct *pp;
  89         struct dtl_entry *dtl;
  90
  91         for_each_possible_cpu(cpu) {
  92                 pp = paca_ptrs[cpu];
  93                 if (pp->dispatch_log)
  94                         continue;
  95                 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
  96                 if (!dtl) {
  97                         pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
  98                                 cpu);
  99 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 100                         pr_warn("Stolen time statistics will be unreliable\n");
 101 #endif
 102                         break;
 103                 }
 104
 105                 pp->dtl_ridx = 0;
 106                 pp->dispatch_log = dtl;
 107                 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
 108                 pp->dtl_curr = dtl;
 109
 110                 if (time_limit && time_after(jiffies, *time_limit)) {
 111                         cond_resched();
 112                         *time_limit = jiffies + HZ;
 113                 }
 114         }
 115 }
 116
 117 void register_dtl_buffer(int cpu)
 118 {
 119         long ret;
 120         struct paca_struct *pp;
 121         struct dtl_entry *dtl;
 122         int hwcpu = get_hard_smp_processor_id(cpu);
 123
 124         pp = paca_ptrs[cpu];
 125         dtl = pp->dispatch_log;
 126         if (dtl && dtl_mask) {
 127                 pp->dtl_ridx = 0;
 128                 pp->dtl_curr = dtl;
 129                 lppaca_of(cpu).dtl_idx = 0;
 130
 131                 /* hypervisor reads buffer length from this field */
 132                 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
 133                 ret = register_dtl(hwcpu, __pa(dtl));
 134                 if (ret)
 135                         pr_err("WARNING: DTL registration of cpu %d (hw %d) failed with %ld\n",
 136                                cpu, hwcpu, ret);
 137
 138                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 139         }
 140 }
 141
 142 #ifdef CONFIG_PPC_SPLPAR
 143 struct dtl_worker {
 144         struct delayed_work work;
 145         int cpu;
 146 };
 147
 148 struct vcpu_dispatch_data {
 149         int last_disp_cpu;
 150
 151         int total_disp;
 152
 153         int same_cpu_disp;
 154         int same_chip_disp;
 155         int diff_chip_disp;
 156         int far_chip_disp;
 157
 158         int numa_home_disp;
 159         int numa_remote_disp;
 160         int numa_far_disp;
 161 };
 162
 163 /*
 164  * This represents the number of cpus in the hypervisor. Since there is no
 165  * architected way to discover the number of processors in the host, we
 166  * provision for dealing with NR_CPUS. This is currently 2048 by default, and
 167  * is sufficient for our purposes. This will need to be tweaked if
 168  * CONFIG_NR_CPUS is changed.
 169  */
 170 #define NR_CPUS_H       NR_CPUS
 171
 172 DEFINE_RWLOCK(dtl_access_lock);
 173 static DEFINE_PER_CPU(struct vcpu_dispatch_data, vcpu_disp_data);
 174 static DEFINE_PER_CPU(u64, dtl_entry_ridx);
 175 static DEFINE_PER_CPU(struct dtl_worker, dtl_workers);
 176 static enum cpuhp_state dtl_worker_state;
 177 static DEFINE_MUTEX(dtl_enable_mutex);
 178 static int vcpudispatch_stats_on __read_mostly;
 179 static int vcpudispatch_stats_freq = 50;
 180 static __be32 *vcpu_associativity, *pcpu_associativity;
 181
 182
 183 static void free_dtl_buffers(unsigned long *time_limit)
 184 {
 185 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 186         int cpu;
 187         struct paca_struct *pp;
 188
 189         for_each_possible_cpu(cpu) {
 190                 pp = paca_ptrs[cpu];
 191                 if (!pp->dispatch_log)
 192                         continue;
 193                 kmem_cache_free(dtl_cache, pp->dispatch_log);
 194                 pp->dtl_ridx = 0;
 195                 pp->dispatch_log = NULL;
 196                 pp->dispatch_log_end = NULL;
 197                 pp->dtl_curr = NULL;
 198
 199                 if (time_limit && time_after(jiffies, *time_limit)) {
 200                         cond_resched();
 201                         *time_limit = jiffies + HZ;
 202                 }
 203         }
 204 #endif
 205 }
 206
 207 static int init_cpu_associativity(void)
 208 {
 209         vcpu_associativity = kcalloc(num_possible_cpus() / threads_per_core,
 210                         VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
 211         pcpu_associativity = kcalloc(NR_CPUS_H / threads_per_core,
 212                         VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
 213
 214         if (!vcpu_associativity || !pcpu_associativity) {
 215                 pr_err("error allocating memory for associativity information\n");
 216                 return -ENOMEM;
 217         }
 218
 219         return 0;
 220 }
 221
 222 static void destroy_cpu_associativity(void)
 223 {
 224         kfree(vcpu_associativity);
 225         kfree(pcpu_associativity);
 226         vcpu_associativity = pcpu_associativity = NULL;
 227 }
 228
 229 static __be32 *__get_cpu_associativity(int cpu, __be32 *cpu_assoc, int flag)
 230 {
 231         __be32 *assoc;
 232         int rc = 0;
 233
 234         assoc = &cpu_assoc[(int)(cpu / threads_per_core) * VPHN_ASSOC_BUFSIZE];
 235         if (!assoc[0]) {
 236                 rc = hcall_vphn(cpu, flag, &assoc[0]);
 237                 if (rc)
 238                         return NULL;
 239         }
 240
 241         return assoc;
 242 }
 243
 244 static __be32 *get_pcpu_associativity(int cpu)
 245 {
 246         return __get_cpu_associativity(cpu, pcpu_associativity, VPHN_FLAG_PCPU);
 247 }
 248
 249 static __be32 *get_vcpu_associativity(int cpu)
 250 {
 251         return __get_cpu_associativity(cpu, vcpu_associativity, VPHN_FLAG_VCPU);
 252 }
 253
 254 static int cpu_relative_dispatch_distance(int last_disp_cpu, int cur_disp_cpu)
 255 {
 256         __be32 *last_disp_cpu_assoc, *cur_disp_cpu_assoc;
 257
 258         if (last_disp_cpu >= NR_CPUS_H || cur_disp_cpu >= NR_CPUS_H)
 259                 return -EINVAL;
 260
 261         last_disp_cpu_assoc = get_pcpu_associativity(last_disp_cpu);
 262         cur_disp_cpu_assoc = get_pcpu_associativity(cur_disp_cpu);
 263
 264         if (!last_disp_cpu_assoc || !cur_disp_cpu_assoc)
 265                 return -EIO;
 266
 267         return cpu_relative_distance(last_disp_cpu_assoc, cur_disp_cpu_assoc);
 268 }
 269
 270 static int cpu_home_node_dispatch_distance(int disp_cpu)
 271 {
 272         __be32 *disp_cpu_assoc, *vcpu_assoc;
 273         int vcpu_id = smp_processor_id();
 274
 275         if (disp_cpu >= NR_CPUS_H) {
 276                 pr_debug_ratelimited("vcpu dispatch cpu %d > %d\n",
 277                                                 disp_cpu, NR_CPUS_H);
 278                 return -EINVAL;
 279         }
 280
 281         disp_cpu_assoc = get_pcpu_associativity(disp_cpu);
 282         vcpu_assoc = get_vcpu_associativity(vcpu_id);
 283
 284         if (!disp_cpu_assoc || !vcpu_assoc)
 285                 return -EIO;
 286
 287         return cpu_relative_distance(disp_cpu_assoc, vcpu_assoc);
 288 }
 289
 290 static void update_vcpu_disp_stat(int disp_cpu)
 291 {
 292         struct vcpu_dispatch_data *disp;
 293         int distance;
 294
 295         disp = this_cpu_ptr(&vcpu_disp_data);
 296         if (disp->last_disp_cpu == -1) {
 297                 disp->last_disp_cpu = disp_cpu;
 298                 return;
 299         }
 300
 301         disp->total_disp++;
 302
 303         if (disp->last_disp_cpu == disp_cpu ||
 304                 (cpu_first_thread_sibling(disp->last_disp_cpu) ==
 305                                         cpu_first_thread_sibling(disp_cpu)))
 306                 disp->same_cpu_disp++;
 307         else {
 308                 distance = cpu_relative_dispatch_distance(disp->last_disp_cpu,
 309                                                                 disp_cpu);
 310                 if (distance < 0)
 311                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
 312                                         smp_processor_id());
 313                 else {
 314                         switch (distance) {
 315                         case 0:
 316                                 disp->same_chip_disp++;
 317                                 break;
 318                         case 1:
 319                                 disp->diff_chip_disp++;
 320                                 break;
 321                         case 2:
 322                                 disp->far_chip_disp++;
 323                                 break;
 324                         default:
 325                                 pr_debug_ratelimited("vcpudispatch_stats: cpu %d (%d -> %d): unexpected relative dispatch distance %d\n",
 326                                                  smp_processor_id(),
 327                                                  disp->last_disp_cpu,
 328                                                  disp_cpu,
 329                                                  distance);
 330                         }
 331                 }
 332         }
 333
 334         distance = cpu_home_node_dispatch_distance(disp_cpu);
 335         if (distance < 0)
 336                 pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
 337                                 smp_processor_id());
 338         else {
 339                 switch (distance) {
 340                 case 0:
 341                         disp->numa_home_disp++;
 342                         break;
 343                 case 1:
 344                         disp->numa_remote_disp++;
 345                         break;
 346                 case 2:
 347                         disp->numa_far_disp++;
 348                         break;
 349                 default:
 350                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d on %d: unexpected numa dispatch distance %d\n",
 351                                                  smp_processor_id(),
 352                                                  disp_cpu,
 353                                                  distance);
 354                 }
 355         }
 356
 357         disp->last_disp_cpu = disp_cpu;
 358 }
 359
 360 static void process_dtl_buffer(struct work_struct *work)
 361 {
 362         struct dtl_entry dtle;
 363         u64 i = __this_cpu_read(dtl_entry_ridx);
 364         struct dtl_entry *dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
 365         struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
 366         struct lppaca *vpa = local_paca->lppaca_ptr;
 367         struct dtl_worker *d = container_of(work, struct dtl_worker, work.work);
 368
 369         if (!local_paca->dispatch_log)
 370                 return;
 371
 372         /* if we have been migrated away, we cancel ourself */
 373         if (d->cpu != smp_processor_id()) {
 374                 pr_debug("vcpudispatch_stats: cpu %d worker migrated -- canceling worker\n",
 375                                                 smp_processor_id());
 376                 return;
 377         }
 378
 379         if (i == be64_to_cpu(vpa->dtl_idx))
 380                 goto out;
 381
 382         while (i < be64_to_cpu(vpa->dtl_idx)) {
 383                 dtle = *dtl;
 384                 barrier();
 385                 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
 386                         /* buffer has overflowed */
 387                         pr_debug_ratelimited("vcpudispatch_stats: cpu %d lost %lld DTL samples\n",
 388                                 d->cpu,
 389                                 be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG - i);
 390                         i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
 391                         dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
 392                         continue;
 393                 }
 394                 update_vcpu_disp_stat(be16_to_cpu(dtle.processor_id));
 395                 ++i;
 396                 ++dtl;
 397                 if (dtl == dtl_end)
 398                         dtl = local_paca->dispatch_log;
 399         }
 400
 401         __this_cpu_write(dtl_entry_ridx, i);
 402
 403 out:
 404         schedule_delayed_work_on(d->cpu, to_delayed_work(work),
 405                                         HZ / vcpudispatch_stats_freq);
 406 }
 407
 408 static int dtl_worker_online(unsigned int cpu)
 409 {
 410         struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
 411
 412         memset(d, 0, sizeof(*d));
 413         INIT_DELAYED_WORK(&d->work, process_dtl_buffer);
 414         d->cpu = cpu;
 415
 416 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 417         per_cpu(dtl_entry_ridx, cpu) = 0;
 418         register_dtl_buffer(cpu);
 419 #else
 420         per_cpu(dtl_entry_ridx, cpu) = be64_to_cpu(lppaca_of(cpu).dtl_idx);
 421 #endif
 422
 423         schedule_delayed_work_on(cpu, &d->work, HZ / vcpudispatch_stats_freq);
 424         return 0;
 425 }
 426
 427 static int dtl_worker_offline(unsigned int cpu)
 428 {
 429         struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
 430
 431         cancel_delayed_work_sync(&d->work);
 432
 433 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 434         unregister_dtl(get_hard_smp_processor_id(cpu));
 435 #endif
 436
 437         return 0;
 438 }
 439
 440 static void set_global_dtl_mask(u8 mask)
 441 {
 442         int cpu;
 443
 444         dtl_mask = mask;
 445         for_each_present_cpu(cpu)
 446                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 447 }
 448
 449 static void reset_global_dtl_mask(void)
 450 {
 451         int cpu;
 452
 453 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 454         dtl_mask = DTL_LOG_PREEMPT;
 455 #else
 456         dtl_mask = 0;
 457 #endif
 458         for_each_present_cpu(cpu)
 459                 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
 460 }
 461
 462 static int dtl_worker_enable(unsigned long *time_limit)
 463 {
 464         int rc = 0, state;
 465
 466         if (!write_trylock(&dtl_access_lock)) {
 467                 rc = -EBUSY;
 468                 goto out;
 469         }
 470
 471         set_global_dtl_mask(DTL_LOG_ALL);
 472
 473         /* Setup dtl buffers and register those */
 474         alloc_dtl_buffers(time_limit);
 475
 476         state = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "powerpc/dtl:online",
 477                                         dtl_worker_online, dtl_worker_offline);
 478         if (state < 0) {
 479                 pr_err("vcpudispatch_stats: unable to setup workqueue for DTL processing\n");
 480                 free_dtl_buffers(time_limit);
 481                 reset_global_dtl_mask();
 482                 write_unlock(&dtl_access_lock);
 483                 rc = -EINVAL;
 484                 goto out;
 485         }
 486         dtl_worker_state = state;
 487
 488 out:
 489         return rc;
 490 }
 491
 492 static void dtl_worker_disable(unsigned long *time_limit)
 493 {
 494         cpuhp_remove_state(dtl_worker_state);
 495         free_dtl_buffers(time_limit);
 496         reset_global_dtl_mask();
 497         write_unlock(&dtl_access_lock);
 498 }
 499
 500 static ssize_t vcpudispatch_stats_write(struct file *file, const char __user *p,
 501                 size_t count, loff_t *ppos)
 502 {
 503         unsigned long time_limit = jiffies + HZ;
 504         struct vcpu_dispatch_data *disp;
 505         int rc, cmd, cpu;
 506         char buf[16];
 507
 508         if (count > 15)
 509                 return -EINVAL;
 510
 511         if (copy_from_user(buf, p, count))
 512                 return -EFAULT;
 513
 514         buf[count] = 0;
 515         rc = kstrtoint(buf, 0, &cmd);
 516         if (rc || cmd < 0 || cmd > 1) {
 517                 pr_err("vcpudispatch_stats: please use 0 to disable or 1 to enable dispatch statistics\n");
 518                 return rc ? rc : -EINVAL;
 519         }
 520
 521         mutex_lock(&dtl_enable_mutex);
 522
 523         if ((cmd == 0 && !vcpudispatch_stats_on) ||
 524                         (cmd == 1 && vcpudispatch_stats_on))
 525                 goto out;
 526
 527         if (cmd) {
 528                 rc = init_cpu_associativity();
 529                 if (rc) {
 530                         destroy_cpu_associativity();
 531                         goto out;
 532                 }
 533
 534                 for_each_possible_cpu(cpu) {
 535                         disp = per_cpu_ptr(&vcpu_disp_data, cpu);
 536                         memset(disp, 0, sizeof(*disp));
 537                         disp->last_disp_cpu = -1;
 538                 }
 539
 540                 rc = dtl_worker_enable(&time_limit);
 541                 if (rc) {
 542                         destroy_cpu_associativity();
 543                         goto out;
 544                 }
 545         } else {
 546                 dtl_worker_disable(&time_limit);
 547                 destroy_cpu_associativity();
 548         }
 549
 550         vcpudispatch_stats_on = cmd;
 551
 552 out:
 553         mutex_unlock(&dtl_enable_mutex);
 554         if (rc)
 555                 return rc;
 556         return count;
 557 }
 558
 559 static int vcpudispatch_stats_display(struct seq_file *p, void *v)
 560 {
 561         int cpu;
 562         struct vcpu_dispatch_data *disp;
 563
 564         if (!vcpudispatch_stats_on) {
 565                 seq_puts(p, "off\n");
 566                 return 0;
 567         }
 568
 569         for_each_online_cpu(cpu) {
 570                 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
 571                 seq_printf(p, "cpu%d", cpu);
 572                 seq_put_decimal_ull(p, " ", disp->total_disp);
 573                 seq_put_decimal_ull(p, " ", disp->same_cpu_disp);
 574                 seq_put_decimal_ull(p, " ", disp->same_chip_disp);
 575                 seq_put_decimal_ull(p, " ", disp->diff_chip_disp);
 576                 seq_put_decimal_ull(p, " ", disp->far_chip_disp);
 577                 seq_put_decimal_ull(p, " ", disp->numa_home_disp);
 578                 seq_put_decimal_ull(p, " ", disp->numa_remote_disp);
 579                 seq_put_decimal_ull(p, " ", disp->numa_far_disp);
 580                 seq_puts(p, "\n");
 581         }
 582
 583         return 0;
 584 }
 585
 586 static int vcpudispatch_stats_open(struct inode *inode, struct file *file)
 587 {
 588         return single_open(file, vcpudispatch_stats_display, NULL);
 589 }
 590
 591 static const struct proc_ops vcpudispatch_stats_proc_ops = {
 592         .proc_open      = vcpudispatch_stats_open,
 593         .proc_read      = seq_read,
 594         .proc_write     = vcpudispatch_stats_write,
 595         .proc_lseek     = seq_lseek,
 596         .proc_release   = single_release,
 597 };
 598
 599 static ssize_t vcpudispatch_stats_freq_write(struct file *file,
 600                 const char __user *p, size_t count, loff_t *ppos)
 601 {
 602         int rc, freq;
 603         char buf[16];
 604
 605         if (count > 15)
 606                 return -EINVAL;
 607
 608         if (copy_from_user(buf, p, count))
 609                 return -EFAULT;
 610
 611         buf[count] = 0;
 612         rc = kstrtoint(buf, 0, &freq);
 613         if (rc || freq < 1 || freq > HZ) {
 614                 pr_err("vcpudispatch_stats_freq: please specify a frequency between 1 and %d\n",
 615                                 HZ);
 616                 return rc ? rc : -EINVAL;
 617         }
 618
 619         vcpudispatch_stats_freq = freq;
 620
 621         return count;
 622 }
 623
 624 static int vcpudispatch_stats_freq_display(struct seq_file *p, void *v)
 625 {
 626         seq_printf(p, "%d\n", vcpudispatch_stats_freq);
 627         return 0;
 628 }
 629
 630 static int vcpudispatch_stats_freq_open(struct inode *inode, struct file *file)
 631 {
 632         return single_open(file, vcpudispatch_stats_freq_display, NULL);
 633 }
 634
 635 static const struct proc_ops vcpudispatch_stats_freq_proc_ops = {
 636         .proc_open      = vcpudispatch_stats_freq_open,
 637         .proc_read      = seq_read,
 638         .proc_write     = vcpudispatch_stats_freq_write,
 639         .proc_lseek     = seq_lseek,
 640         .proc_release   = single_release,
 641 };
 642
 643 static int __init vcpudispatch_stats_procfs_init(void)
 644 {
 645         if (!lppaca_shared_proc())
 646                 return 0;
 647
 648         if (!proc_create("powerpc/vcpudispatch_stats", 0600, NULL,
 649                                         &vcpudispatch_stats_proc_ops))
 650                 pr_err("vcpudispatch_stats: error creating procfs file\n");
 651         else if (!proc_create("powerpc/vcpudispatch_stats_freq", 0600, NULL,
 652                                         &vcpudispatch_stats_freq_proc_ops))
 653                 pr_err("vcpudispatch_stats_freq: error creating procfs file\n");
 654
 655         return 0;
 656 }
 657
 658 machine_device_initcall(pseries, vcpudispatch_stats_procfs_init);
 659
 660 #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
 661 u64 pseries_paravirt_steal_clock(int cpu)
 662 {
 663         struct lppaca *lppaca = &lppaca_of(cpu);
 664
 665         return be64_to_cpu(READ_ONCE(lppaca->enqueue_dispatch_tb)) +
 666                 be64_to_cpu(READ_ONCE(lppaca->ready_enqueue_tb));
 667 }
 668 #endif
 669
 670 #endif /* CONFIG_PPC_SPLPAR */
 671
 672 void vpa_init(int cpu)
 673 {
 674         int hwcpu = get_hard_smp_processor_id(cpu);
 675         unsigned long addr;
 676         long ret;
 677
 678         /*
 679          * The spec says it "may be problematic" if CPU x registers the VPA of
 680          * CPU y. We should never do that, but wail if we ever do.
 681          */
 682         WARN_ON(cpu != smp_processor_id());
 683
 684         if (cpu_has_feature(CPU_FTR_ALTIVEC))
 685                 lppaca_of(cpu).vmxregs_in_use = 1;
 686
 687         if (cpu_has_feature(CPU_FTR_ARCH_207S))
 688                 lppaca_of(cpu).ebb_regs_in_use = 1;
 689
 690         addr = __pa(&lppaca_of(cpu));
 691         ret = register_vpa(hwcpu, addr);
 692
 693         if (ret) {
 694                 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
 695                        "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
 696                 return;
 697         }
 698
 699 #ifdef CONFIG_PPC_64S_HASH_MMU
 700         /*
 701          * PAPR says this feature is SLB-Buffer but firmware never
 702          * reports that.  All SPLPAR support SLB shadow buffer.
 703          */
 704         if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
 705                 addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
 706                 ret = register_slb_shadow(hwcpu, addr);
 707                 if (ret)
 708                         pr_err("WARNING: SLB shadow buffer registration for "
 709                                "cpu %d (hw %d) of area %lx failed with %ld\n",
 710                                cpu, hwcpu, addr, ret);
 711         }
 712 #endif /* CONFIG_PPC_64S_HASH_MMU */
 713
 714         /*
 715          * Register dispatch trace log, if one has been allocated.
 716          */
 717         register_dtl_buffer(cpu);
 718 }
 719
 720 #ifdef CONFIG_PPC_BOOK3S_64
 721
 722 static int __init pseries_lpar_register_process_table(unsigned long base,
 723                         unsigned long page_size, unsigned long table_size)
 724 {
 725         long rc;
 726         unsigned long flags = 0;
 727
 728         if (table_size)
 729                 flags |= PROC_TABLE_NEW;
 730         if (radix_enabled()) {
 731                 flags |= PROC_TABLE_RADIX;
 732                 if (mmu_has_feature(MMU_FTR_GTSE))
 733                         flags |= PROC_TABLE_GTSE;
 734         } else
 735                 flags |= PROC_TABLE_HPT_SLB;
 736         for (;;) {
 737                 rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
 738                                         page_size, table_size);
 739                 if (!H_IS_LONG_BUSY(rc))
 740                         break;
 741                 mdelay(get_longbusy_msecs(rc));
 742         }
 743         if (rc != H_SUCCESS) {
 744                 pr_err("Failed to register process table (rc=%ld)\n", rc);
 745                 BUG();
 746         }
 747         return rc;
 748 }
 749
 750 #ifdef CONFIG_PPC_64S_HASH_MMU
 751
 752 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 753                                      unsigned long vpn, unsigned long pa,
 754                                      unsigned long rflags, unsigned long vflags,
 755                                      int psize, int apsize, int ssize)
 756 {
 757         unsigned long lpar_rc;
 758         unsigned long flags;
 759         unsigned long slot;
 760         unsigned long hpte_v, hpte_r;
 761
 762         if (!(vflags & HPTE_V_BOLTED))
 763                 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
 764                          "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
 765                          hpte_group, vpn,  pa, rflags, vflags, psize);
 766
 767         hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
 768         hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
 769
 770         if (!(vflags & HPTE_V_BOLTED))
 771                 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 772
 773         /* Now fill in the actual HPTE */
 774         /* Set CEC cookie to 0         */
 775         /* Zero page = 0               */
 776         /* I-cache Invalidate = 0      */
 777         /* I-cache synchronize = 0     */
 778         /* Exact = 0                   */
 779         flags = 0;
 780
 781         if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
 782                 flags |= H_COALESCE_CAND;
 783
 784         lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
 785         if (unlikely(lpar_rc == H_PTEG_FULL)) {
 786                 pr_devel("Hash table group is full\n");
 787                 return -1;
 788         }
 789
 790         /*
 791          * Since we try and ioremap PHBs we don't own, the pte insert
 792          * will fail. However we must catch the failure in hash_page
 793          * or we will loop forever, so return -2 in this case.
 794          */
 795         if (unlikely(lpar_rc != H_SUCCESS)) {
 796                 pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
 797                 return -2;
 798         }
 799         if (!(vflags & HPTE_V_BOLTED))
 800                 pr_devel(" -> slot: %lu\n", slot & 7);
 801
 802         /* Because of iSeries, we have to pass down the secondary
 803          * bucket bit here as well
 804          */
 805         return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
 806 }
 807
 808 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
 809
 810 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
 811 {
 812         unsigned long slot_offset;
 813         unsigned long lpar_rc;
 814         int i;
 815         unsigned long dummy1, dummy2;
 816
 817         /* pick a random slot to start at */
 818         slot_offset = mftb() & 0x7;
 819
 820         for (i = 0; i < HPTES_PER_GROUP; i++) {
 821
 822                 /* don't remove a bolted entry */
 823                 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
 824                                            HPTE_V_BOLTED, &dummy1, &dummy2);
 825                 if (lpar_rc == H_SUCCESS)
 826                         return i;
 827
 828                 /*
 829                  * The test for adjunct partition is performed before the
 830                  * ANDCOND test.  H_RESOURCE may be returned, so we need to
 831                  * check for that as well.
 832                  */
 833                 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
 834
 835                 slot_offset++;
 836                 slot_offset &= 0x7;
 837         }
 838
 839         return -1;
 840 }
 841
 842 /* Called during kexec sequence with MMU off */
 843 static notrace void manual_hpte_clear_all(void)
 844 {
 845         unsigned long size_bytes = 1UL << ppc64_pft_size;
 846         unsigned long hpte_count = size_bytes >> 4;
 847         struct {
 848                 unsigned long pteh;
 849                 unsigned long ptel;
 850         } ptes[4];
 851         long lpar_rc;
 852         unsigned long i, j;
 853
 854         /* Read in batches of 4,
 855          * invalidate only valid entries not in the VRMA
 856          * hpte_count will be a multiple of 4
 857          */
 858         for (i = 0; i < hpte_count; i += 4) {
 859                 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
 860                 if (lpar_rc != H_SUCCESS) {
 861                         pr_info("Failed to read hash page table at %ld err %ld\n",
 862                                 i, lpar_rc);
 863                         continue;
 864                 }
 865                 for (j = 0; j < 4; j++){
 866                         if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
 867                                 HPTE_V_VRMA_MASK)
 868                                 continue;
 869                         if (ptes[j].pteh & HPTE_V_VALID)
 870                                 plpar_pte_remove_raw(0, i + j, 0,
 871                                         &(ptes[j].pteh), &(ptes[j].ptel));
 872                 }
 873         }
 874 }
 875
 876 /* Called during kexec sequence with MMU off */
 877 static notrace int hcall_hpte_clear_all(void)
 878 {
 879         int rc;
 880
 881         do {
 882                 rc = plpar_hcall_norets(H_CLEAR_HPT);
 883         } while (rc == H_CONTINUE);
 884
 885         return rc;
 886 }
 887
 888 /* Called during kexec sequence with MMU off */
 889 static notrace void pseries_hpte_clear_all(void)
 890 {
 891         int rc;
 892
 893         rc = hcall_hpte_clear_all();
 894         if (rc != H_SUCCESS)
 895                 manual_hpte_clear_all();
 896
 897 #ifdef __LITTLE_ENDIAN__
 898         /*
 899          * Reset exceptions to big endian.
 900          *
 901          * FIXME this is a hack for kexec, we need to reset the exception
 902          * endian before starting the new kernel and this is a convenient place
 903          * to do it.
 904          *
 905          * This is also called on boot when a fadump happens. In that case we
 906          * must not change the exception endian mode.
 907          */
 908         if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
 909                 pseries_big_endian_exceptions();
 910 #endif
 911 }
 912
 913 /*
 914  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 915  * the low 3 bits of flags happen to line up.  So no transform is needed.
 916  * We can probably optimize here and assume the high bits of newpp are
 917  * already zero.  For now I am paranoid.
 918  */
 919 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 920                                        unsigned long newpp,
 921                                        unsigned long vpn,
 922                                        int psize, int apsize,
 923                                        int ssize, unsigned long inv_flags)
 924 {
 925         unsigned long lpar_rc;
 926         unsigned long flags;
 927         unsigned long want_v;
 928
 929         want_v = hpte_encode_avpn(vpn, psize, ssize);
 930
 931         flags = (newpp & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO)) | H_AVPN;
 932         flags |= (newpp & HPTE_R_KEY_HI) >> 48;
 933         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
 934                 /* Move pp0 into bit 8 (IBM 55) */
 935                 flags |= (newpp & HPTE_R_PP0) >> 55;
 936
 937         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 938                  want_v, slot, flags, psize);
 939
 940         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 941
 942         if (lpar_rc == H_NOT_FOUND) {
 943                 pr_devel("not found !\n");
 944                 return -1;
 945         }
 946
 947         pr_devel("ok\n");
 948
 949         BUG_ON(lpar_rc != H_SUCCESS);
 950
 951         return 0;
 952 }
 953
 954 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
 955 {
 956         long lpar_rc;
 957         unsigned long i, j;
 958         struct {
 959                 unsigned long pteh;
 960                 unsigned long ptel;
 961         } ptes[4];
 962
 963         for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
 964
 965                 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
 966                 if (lpar_rc != H_SUCCESS) {
 967                         pr_info("Failed to read hash page table at %ld err %ld\n",
 968                                 hpte_group, lpar_rc);
 969                         continue;
 970                 }
 971
 972                 for (j = 0; j < 4; j++) {
 973                         if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
 974                             (ptes[j].pteh & HPTE_V_VALID))
 975                                 return i + j;
 976                 }
 977         }
 978
 979         return -1;
 980 }
 981
 982 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
 983 {
 984         long slot;
 985         unsigned long hash;
 986         unsigned long want_v;
 987         unsigned long hpte_group;
 988
 989         hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
 990         want_v = hpte_encode_avpn(vpn, psize, ssize);
 991
 992         /*
 993          * We try to keep bolted entries always in primary hash
 994          * But in some case we can find them in secondary too.
 995          */
 996         hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 997         slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
 998         if (slot < 0) {
 999                 /* Try in secondary */
1000                 hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
1001                 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
1002                 if (slot < 0)
1003                         return -1;
1004         }
1005         return hpte_group + slot;
1006 }
1007
1008 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
1009                                              unsigned long ea,
1010                                              int psize, int ssize)
1011 {
1012         unsigned long vpn;
1013         unsigned long lpar_rc, slot, vsid, flags;
1014
1015         vsid = get_kernel_vsid(ea, ssize);
1016         vpn = hpt_vpn(ea, vsid, ssize);
1017
1018         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1019         BUG_ON(slot == -1);
1020
1021         flags = newpp & (HPTE_R_PP | HPTE_R_N);
1022         if (mmu_has_feature(MMU_FTR_KERNEL_RO))
1023                 /* Move pp0 into bit 8 (IBM 55) */
1024                 flags |= (newpp & HPTE_R_PP0) >> 55;
1025
1026         flags |= ((newpp & HPTE_R_KEY_HI) >> 48) | (newpp & HPTE_R_KEY_LO);
1027
1028         lpar_rc = plpar_pte_protect(flags, slot, 0);
1029
1030         BUG_ON(lpar_rc != H_SUCCESS);
1031 }
1032
1033 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
1034                                          int psize, int apsize,
1035                                          int ssize, int local)
1036 {
1037         unsigned long want_v;
1038         unsigned long lpar_rc;
1039         unsigned long dummy1, dummy2;
1040
1041         pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
1042                  slot, vpn, psize, local);
1043
1044         want_v = hpte_encode_avpn(vpn, psize, ssize);
1045         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
1046         if (lpar_rc == H_NOT_FOUND)
1047                 return;
1048
1049         BUG_ON(lpar_rc != H_SUCCESS);
1050 }
1051
1052
1053 /*
1054  * As defined in the PAPR's section 14.5.4.1.8
1055  * The control mask doesn't include the returned reference and change bit from
1056  * the processed PTE.
1057  */
1058 #define HBLKR_AVPN              0x0100000000000000UL
1059 #define HBLKR_CTRL_MASK         0xf800000000000000UL
1060 #define HBLKR_CTRL_SUCCESS      0x8000000000000000UL
1061 #define HBLKR_CTRL_ERRNOTFOUND  0x8800000000000000UL
1062 #define HBLKR_CTRL_ERRBUSY      0xa000000000000000UL
1063
1064 /*
1065  * Returned true if we are supporting this block size for the specified segment
1066  * base page size and actual page size.
1067  *
1068  * Currently, we only support 8 size block.
1069  */
1070 static inline bool is_supported_hlbkrm(int bpsize, int psize)
1071 {
1072         return (hblkrm_size[bpsize][psize] == HBLKRM_SUPPORTED_BLOCK_SIZE);
1073 }
1074
1075 /**
1076  * H_BLOCK_REMOVE caller.
1077  * @idx should point to the latest @param entry set with a PTEX.
1078  * If PTE cannot be processed because another CPUs has already locked that
1079  * group, those entries are put back in @param starting at index 1.
1080  * If entries has to be retried and @retry_busy is set to true, these entries
1081  * are retried until success. If @retry_busy is set to false, the returned
1082  * is the number of entries yet to process.
1083  */
1084 static unsigned long call_block_remove(unsigned long idx, unsigned long *param,
1085                                        bool retry_busy)
1086 {
1087         unsigned long i, rc, new_idx;
1088         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1089
1090         if (idx < 2) {
1091                 pr_warn("Unexpected empty call to H_BLOCK_REMOVE");
1092                 return 0;
1093         }
1094 again:
1095         new_idx = 0;
1096         if (idx > PLPAR_HCALL9_BUFSIZE) {
1097                 pr_err("Too many PTEs (%lu) for H_BLOCK_REMOVE", idx);
1098                 idx = PLPAR_HCALL9_BUFSIZE;
1099         } else if (idx < PLPAR_HCALL9_BUFSIZE)
1100                 param[idx] = HBR_END;
1101
1102         rc = plpar_hcall9(H_BLOCK_REMOVE, retbuf,
1103                           param[0], /* AVA */
1104                           param[1],  param[2],  param[3],  param[4], /* TS0-7 */
1105                           param[5],  param[6],  param[7],  param[8]);
1106         if (rc == H_SUCCESS)
1107                 return 0;
1108
1109         BUG_ON(rc != H_PARTIAL);
1110
1111         /* Check that the unprocessed entries were 'not found' or 'busy' */
1112         for (i = 0; i < idx-1; i++) {
1113                 unsigned long ctrl = retbuf[i] & HBLKR_CTRL_MASK;
1114
1115                 if (ctrl == HBLKR_CTRL_ERRBUSY) {
1116                         param[++new_idx] = param[i+1];
1117                         continue;
1118                 }
1119
1120                 BUG_ON(ctrl != HBLKR_CTRL_SUCCESS
1121                        && ctrl != HBLKR_CTRL_ERRNOTFOUND);
1122         }
1123
1124         /*
1125          * If there were entries found busy, retry these entries if requested,
1126          * of if all the entries have to be retried.
1127          */
1128         if (new_idx && (retry_busy || new_idx == (PLPAR_HCALL9_BUFSIZE-1))) {
1129                 idx = new_idx + 1;
1130                 goto again;
1131         }
1132
1133         return new_idx;
1134 }
1135
1136 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1137 /*
1138  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
1139  * to make sure that we avoid bouncing the hypervisor tlbie lock.
1140  */
1141 #define PPC64_HUGE_HPTE_BATCH 12
1142
1143 static void hugepage_block_invalidate(unsigned long *slot, unsigned long *vpn,
1144                                       int count, int psize, int ssize)
1145 {
1146         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1147         unsigned long shift, current_vpgb, vpgb;
1148         int i, pix = 0;
1149
1150         shift = mmu_psize_defs[psize].shift;
1151
1152         for (i = 0; i < count; i++) {
1153                 /*
1154                  * Shifting 3 bits more on the right to get a
1155                  * 8 pages aligned virtual addresse.
1156                  */
1157                 vpgb = (vpn[i] >> (shift - VPN_SHIFT + 3));
1158                 if (!pix || vpgb != current_vpgb) {
1159                         /*
1160                          * Need to start a new 8 pages block, flush
1161                          * the current one if needed.
1162                          */
1163                         if (pix)
1164                                 (void)call_block_remove(pix, param, true);
1165                         current_vpgb = vpgb;
1166                         param[0] = hpte_encode_avpn(vpn[i], psize, ssize);
1167                         pix = 1;
1168                 }
1169
1170                 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot[i];
1171                 if (pix == PLPAR_HCALL9_BUFSIZE) {
1172                         pix = call_block_remove(pix, param, false);
1173                         /*
1174                          * pix = 0 means that all the entries were
1175                          * removed, we can start a new block.
1176                          * Otherwise, this means that there are entries
1177                          * to retry, and pix points to latest one, so
1178                          * we should increment it and try to continue
1179                          * the same block.
1180                          */
1181                         if (pix)
1182                                 pix++;
1183                 }
1184         }
1185         if (pix)
1186                 (void)call_block_remove(pix, param, true);
1187 }
1188
1189 static void hugepage_bulk_invalidate(unsigned long *slot, unsigned long *vpn,
1190                                      int count, int psize, int ssize)
1191 {
1192         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1193         int i = 0, pix = 0, rc;
1194
1195         for (i = 0; i < count; i++) {
1196
1197                 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1198                         pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
1199                                                      ssize, 0);
1200                 } else {
1201                         param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
1202                         param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
1203                         pix += 2;
1204                         if (pix == 8) {
1205                                 rc = plpar_hcall9(H_BULK_REMOVE, param,
1206                                                   param[0], param[1], param[2],
1207                                                   param[3], param[4], param[5],
1208                                                   param[6], param[7]);
1209                                 BUG_ON(rc != H_SUCCESS);
1210                                 pix = 0;
1211                         }
1212                 }
1213         }
1214         if (pix) {
1215                 param[pix] = HBR_END;
1216                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1217                                   param[2], param[3], param[4], param[5],
1218                                   param[6], param[7]);
1219                 BUG_ON(rc != H_SUCCESS);
1220         }
1221 }
1222
1223 static inline void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
1224                                                       unsigned long *vpn,
1225                                                       int count, int psize,
1226                                                       int ssize)
1227 {
1228         unsigned long flags = 0;
1229         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1230
1231         if (lock_tlbie)
1232                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1233
1234         /* Assuming THP size is 16M */
1235         if (is_supported_hlbkrm(psize, MMU_PAGE_16M))
1236                 hugepage_block_invalidate(slot, vpn, count, psize, ssize);
1237         else
1238                 hugepage_bulk_invalidate(slot, vpn, count, psize, ssize);
1239
1240         if (lock_tlbie)
1241                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1242 }
1243
1244 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1245                                              unsigned long addr,
1246                                              unsigned char *hpte_slot_array,
1247                                              int psize, int ssize, int local)
1248 {
1249         int i, index = 0;
1250         unsigned long s_addr = addr;
1251         unsigned int max_hpte_count, valid;
1252         unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
1253         unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
1254         unsigned long shift, hidx, vpn = 0, hash, slot;
1255
1256         shift = mmu_psize_defs[psize].shift;
1257         max_hpte_count = 1U << (PMD_SHIFT - shift);
1258
1259         for (i = 0; i < max_hpte_count; i++) {
1260                 valid = hpte_valid(hpte_slot_array, i);
1261                 if (!valid)
1262                         continue;
1263                 hidx =  hpte_hash_index(hpte_slot_array, i);
1264
1265                 /* get the vpn */
1266                 addr = s_addr + (i * (1ul << shift));
1267                 vpn = hpt_vpn(addr, vsid, ssize);
1268                 hash = hpt_hash(vpn, shift, ssize);
1269                 if (hidx & _PTEIDX_SECONDARY)
1270                         hash = ~hash;
1271
1272                 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1273                 slot += hidx & _PTEIDX_GROUP_IX;
1274
1275                 slot_array[index] = slot;
1276                 vpn_array[index] = vpn;
1277                 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
1278                         /*
1279                          * Now do a bluk invalidate
1280                          */
1281                         __pSeries_lpar_hugepage_invalidate(slot_array,
1282                                                            vpn_array,
1283                                                            PPC64_HUGE_HPTE_BATCH,
1284                                                            psize, ssize);
1285                         index = 0;
1286                 } else
1287                         index++;
1288         }
1289         if (index)
1290                 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
1291                                                    index, psize, ssize);
1292 }
1293 #else
1294 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1295                                              unsigned long addr,
1296                                              unsigned char *hpte_slot_array,
1297                                              int psize, int ssize, int local)
1298 {
1299         WARN(1, "%s called without THP support\n", __func__);
1300 }
1301 #endif
1302
1303 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
1304                                           int psize, int ssize)
1305 {
1306         unsigned long vpn;
1307         unsigned long slot, vsid;
1308
1309         vsid = get_kernel_vsid(ea, ssize);
1310         vpn = hpt_vpn(ea, vsid, ssize);
1311
1312         slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1313         if (slot == -1)
1314                 return -ENOENT;
1315
1316         /*
1317          * lpar doesn't use the passed actual page size
1318          */
1319         pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
1320         return 0;
1321 }
1322
1323
1324 static inline unsigned long compute_slot(real_pte_t pte,
1325                                          unsigned long vpn,
1326                                          unsigned long index,
1327                                          unsigned long shift,
1328                                          int ssize)
1329 {
1330         unsigned long slot, hash, hidx;
1331
1332         hash = hpt_hash(vpn, shift, ssize);
1333         hidx = __rpte_to_hidx(pte, index);
1334         if (hidx & _PTEIDX_SECONDARY)
1335                 hash = ~hash;
1336         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1337         slot += hidx & _PTEIDX_GROUP_IX;
1338         return slot;
1339 }
1340
1341 /**
1342  * The hcall H_BLOCK_REMOVE implies that the virtual pages to processed are
1343  * "all within the same naturally aligned 8 page virtual address block".
1344  */
1345 static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
1346                             unsigned long *param)
1347 {
1348         unsigned long vpn;
1349         unsigned long i, pix = 0;
1350         unsigned long index, shift, slot, current_vpgb, vpgb;
1351         real_pte_t pte;
1352         int psize, ssize;
1353
1354         psize = batch->psize;
1355         ssize = batch->ssize;
1356
1357         for (i = 0; i < number; i++) {
1358                 vpn = batch->vpn[i];
1359                 pte = batch->pte[i];
1360                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1361                         /*
1362                          * Shifting 3 bits more on the right to get a
1363                          * 8 pages aligned virtual addresse.
1364                          */
1365                         vpgb = (vpn >> (shift - VPN_SHIFT + 3));
1366                         if (!pix || vpgb != current_vpgb) {
1367                                 /*
1368                                  * Need to start a new 8 pages block, flush
1369                                  * the current one if needed.
1370                                  */
1371                                 if (pix)
1372                                         (void)call_block_remove(pix, param,
1373                                                                 true);
1374                                 current_vpgb = vpgb;
1375                                 param[0] = hpte_encode_avpn(vpn, psize,
1376                                                             ssize);
1377                                 pix = 1;
1378                         }
1379
1380                         slot = compute_slot(pte, vpn, index, shift, ssize);
1381                         param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot;
1382
1383                         if (pix == PLPAR_HCALL9_BUFSIZE) {
1384                                 pix = call_block_remove(pix, param, false);
1385                                 /*
1386                                  * pix = 0 means that all the entries were
1387                                  * removed, we can start a new block.
1388                                  * Otherwise, this means that there are entries
1389                                  * to retry, and pix points to latest one, so
1390                                  * we should increment it and try to continue
1391                                  * the same block.
1392                                  */
1393                                 if (pix)
1394                                         pix++;
1395                         }
1396                 } pte_iterate_hashed_end();
1397         }
1398
1399         if (pix)
1400                 (void)call_block_remove(pix, param, true);
1401 }
1402
1403 /*
1404  * TLB Block Invalidate Characteristics
1405  *
1406  * These characteristics define the size of the block the hcall H_BLOCK_REMOVE
1407  * is able to process for each couple segment base page size, actual page size.
1408  *
1409  * The ibm,get-system-parameter properties is returning a buffer with the
1410  * following layout:
1411  *
1412  * [ 2 bytes size of the RTAS buffer (excluding these 2 bytes) ]
1413  * -----------------
1414  * TLB Block Invalidate Specifiers:
1415  * [ 1 byte LOG base 2 of the TLB invalidate block size being specified ]
1416  * [ 1 byte Number of page sizes (N) that are supported for the specified
1417  *          TLB invalidate block size ]
1418  * [ 1 byte Encoded segment base page size and actual page size
1419  *          MSB=0 means 4k segment base page size and actual page size
1420  *          MSB=1 the penc value in mmu_psize_def ]
1421  * ...
1422  * -----------------
1423  * Next TLB Block Invalidate Specifiers...
1424  * -----------------
1425  * [ 0 ]
1426  */
1427 static inline void set_hblkrm_bloc_size(int bpsize, int psize,
1428                                         unsigned int block_size)
1429 {
1430         if (block_size > hblkrm_size[bpsize][psize])
1431                 hblkrm_size[bpsize][psize] = block_size;
1432 }
1433
1434 /*
1435  * Decode the Encoded segment base page size and actual page size.
1436  * PAPR specifies:
1437  *   - bit 7 is the L bit
1438  *   - bits 0-5 are the penc value
1439  * If the L bit is 0, this means 4K segment base page size and actual page size
1440  * otherwise the penc value should be read.
1441  */
1442 #define HBLKRM_L_MASK           0x80
1443 #define HBLKRM_PENC_MASK        0x3f
1444 static inline void __init check_lp_set_hblkrm(unsigned int lp,
1445                                               unsigned int block_size)
1446 {
1447         unsigned int bpsize, psize;
1448
1449         /* First, check the L bit, if not set, this means 4K */
1450         if ((lp & HBLKRM_L_MASK) == 0) {
1451                 set_hblkrm_bloc_size(MMU_PAGE_4K, MMU_PAGE_4K, block_size);
1452                 return;
1453         }
1454
1455         lp &= HBLKRM_PENC_MASK;
1456         for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) {
1457                 struct mmu_psize_def *def = &mmu_psize_defs[bpsize];
1458
1459                 for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1460                         if (def->penc[psize] == lp) {
1461                                 set_hblkrm_bloc_size(bpsize, psize, block_size);
1462                                 return;
1463                         }
1464                 }
1465         }
1466 }
1467
1468 /*
1469  * The size of the TLB Block Invalidate Characteristics is variable. But at the
1470  * maximum it will be the number of possible page sizes *2 + 10 bytes.
1471  * Currently MMU_PAGE_COUNT is 16, which means 42 bytes. Use a cache line size
1472  * (128 bytes) for the buffer to get plenty of space.
1473  */
1474 #define SPLPAR_TLB_BIC_MAXLENGTH        128
1475
1476 void __init pseries_lpar_read_hblkrm_characteristics(void)
1477 {
1478         static struct papr_sysparm_buf buf __initdata;
1479         int len, idx, bpsize;
1480
1481         if (!firmware_has_feature(FW_FEATURE_BLOCK_REMOVE))
1482                 return;
1483
1484         if (papr_sysparm_get(PAPR_SYSPARM_TLB_BLOCK_INVALIDATE_ATTRS, &buf))
1485                 return;
1486
1487         len = be16_to_cpu(buf.len);
1488         if (len > SPLPAR_TLB_BIC_MAXLENGTH) {
1489                 pr_warn("%s too large returned buffer %d", __func__, len);
1490                 return;
1491         }
1492
1493         idx = 0;
1494         while (idx < len) {
1495                 u8 block_shift = buf.val[idx++];
1496                 u32 block_size;
1497                 unsigned int npsize;
1498
1499                 if (!block_shift)
1500                         break;
1501
1502                 block_size = 1 << block_shift;
1503
1504                 for (npsize = buf.val[idx++];
1505                      npsize > 0 && idx < len; npsize--)
1506                         check_lp_set_hblkrm((unsigned int)buf.val[idx++],
1507                                             block_size);
1508         }
1509
1510         for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
1511                 for (idx = 0; idx < MMU_PAGE_COUNT; idx++)
1512                         if (hblkrm_size[bpsize][idx])
1513                                 pr_info("H_BLOCK_REMOVE supports base psize:%d psize:%d block size:%d",
1514                                         bpsize, idx, hblkrm_size[bpsize][idx]);
1515 }
1516
1517 /*
1518  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
1519  * lock.
1520  */
1521 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
1522 {
1523         unsigned long vpn;
1524         unsigned long i, pix, rc;
1525         unsigned long flags = 0;
1526         struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
1527         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1528         unsigned long param[PLPAR_HCALL9_BUFSIZE];
1529         unsigned long index, shift, slot;
1530         real_pte_t pte;
1531         int psize, ssize;
1532
1533         if (lock_tlbie)
1534                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1535
1536         if (is_supported_hlbkrm(batch->psize, batch->psize)) {
1537                 do_block_remove(number, batch, param);
1538                 goto out;
1539         }
1540
1541         psize = batch->psize;
1542         ssize = batch->ssize;
1543         pix = 0;
1544         for (i = 0; i < number; i++) {
1545                 vpn = batch->vpn[i];
1546                 pte = batch->pte[i];
1547                 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1548                         slot = compute_slot(pte, vpn, index, shift, ssize);
1549                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1550                                 /*
1551                                  * lpar doesn't use the passed actual page size
1552                                  */
1553                                 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
1554                                                              0, ssize, local);
1555                         } else {
1556                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
1557                                 param[pix+1] = hpte_encode_avpn(vpn, psize,
1558                                                                 ssize);
1559                                 pix += 2;
1560                                 if (pix == 8) {
1561                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
1562                                                 param[0], param[1], param[2],
1563                                                 param[3], param[4], param[5],
1564                                                 param[6], param[7]);
1565                                         BUG_ON(rc != H_SUCCESS);
1566                                         pix = 0;
1567                                 }
1568                         }
1569                 } pte_iterate_hashed_end();
1570         }
1571         if (pix) {
1572                 param[pix] = HBR_END;
1573                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1574                                   param[2], param[3], param[4], param[5],
1575                                   param[6], param[7]);
1576                 BUG_ON(rc != H_SUCCESS);
1577         }
1578
1579 out:
1580         if (lock_tlbie)
1581                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1582 }
1583
1584 static int __init disable_bulk_remove(char *str)
1585 {
1586         if (strcmp(str, "off") == 0 &&
1587             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1588                 pr_info("Disabling BULK_REMOVE firmware feature");
1589                 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
1590         }
1591         return 1;
1592 }
1593
1594 __setup("bulk_remove=", disable_bulk_remove);
1595
1596 #define HPT_RESIZE_TIMEOUT      10000 /* ms */
1597
1598 struct hpt_resize_state {
1599         unsigned long shift;
1600         int commit_rc;
1601 };
1602
1603 static int pseries_lpar_resize_hpt_commit(void *data)
1604 {
1605         struct hpt_resize_state *state = data;
1606
1607         state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
1608         if (state->commit_rc != H_SUCCESS)
1609                 return -EIO;
1610
1611         /* Hypervisor has transitioned the HTAB, update our globals */
1612         ppc64_pft_size = state->shift;
1613         htab_size_bytes = 1UL << ppc64_pft_size;
1614         htab_hash_mask = (htab_size_bytes >> 7) - 1;
1615
1616         return 0;
1617 }
1618
1619 /*
1620  * Must be called in process context. The caller must hold the
1621  * cpus_lock.
1622  */
1623 static int pseries_lpar_resize_hpt(unsigned long shift)
1624 {
1625         struct hpt_resize_state state = {
1626                 .shift = shift,
1627                 .commit_rc = H_FUNCTION,
1628         };
1629         unsigned int delay, total_delay = 0;
1630         int rc;
1631         ktime_t t0, t1, t2;
1632
1633         might_sleep();
1634
1635         if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1636                 return -ENODEV;
1637
1638         pr_info("Attempting to resize HPT to shift %lu\n", shift);
1639
1640         t0 = ktime_get();
1641
1642         rc = plpar_resize_hpt_prepare(0, shift);
1643         while (H_IS_LONG_BUSY(rc)) {
1644                 delay = get_longbusy_msecs(rc);
1645                 total_delay += delay;
1646                 if (total_delay > HPT_RESIZE_TIMEOUT) {
1647                         /* prepare with shift==0 cancels an in-progress resize */
1648                         rc = plpar_resize_hpt_prepare(0, 0);
1649                         if (rc != H_SUCCESS)
1650                                 pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
1651                                        rc);
1652                         return -ETIMEDOUT;
1653                 }
1654                 msleep(delay);
1655                 rc = plpar_resize_hpt_prepare(0, shift);
1656         }
1657
1658         switch (rc) {
1659         case H_SUCCESS:
1660                 /* Continue on */
1661                 break;
1662
1663         case H_PARAMETER:
1664                 pr_warn("Invalid argument from H_RESIZE_HPT_PREPARE\n");
1665                 return -EINVAL;
1666         case H_RESOURCE:
1667                 pr_warn("Operation not permitted from H_RESIZE_HPT_PREPARE\n");
1668                 return -EPERM;
1669         default:
1670                 pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
1671                 return -EIO;
1672         }
1673
1674         t1 = ktime_get();
1675
1676         rc = stop_machine_cpuslocked(pseries_lpar_resize_hpt_commit,
1677                                      &state, NULL);
1678
1679         t2 = ktime_get();
1680
1681         if (rc != 0) {
1682                 switch (state.commit_rc) {
1683                 case H_PTEG_FULL:
1684                         return -ENOSPC;
1685
1686                 default:
1687                         pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
1688                                 state.commit_rc);
1689                         return -EIO;
1690                 };
1691         }
1692
1693         pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
1694                 shift, (long long) ktime_ms_delta(t1, t0),
1695                 (long long) ktime_ms_delta(t2, t1));
1696
1697         return 0;
1698 }
1699
1700 void __init hpte_init_pseries(void)
1701 {
1702         mmu_hash_ops.hpte_invalidate     = pSeries_lpar_hpte_invalidate;
1703         mmu_hash_ops.hpte_updatepp       = pSeries_lpar_hpte_updatepp;
1704         mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
1705         mmu_hash_ops.hpte_insert         = pSeries_lpar_hpte_insert;
1706         mmu_hash_ops.hpte_remove         = pSeries_lpar_hpte_remove;
1707         mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
1708         mmu_hash_ops.flush_hash_range    = pSeries_lpar_flush_hash_range;
1709         mmu_hash_ops.hpte_clear_all      = pseries_hpte_clear_all;
1710         mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
1711
1712         if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1713                 mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
1714
1715         /*
1716          * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall
1717          * to inform the hypervisor that we wish to use the HPT.
1718          */
1719         if (cpu_has_feature(CPU_FTR_ARCH_300))
1720                 pseries_lpar_register_process_table(0, 0, 0);
1721 }
1722 #endif /* CONFIG_PPC_64S_HASH_MMU */
1723
1724 #ifdef CONFIG_PPC_RADIX_MMU
1725 void __init radix_init_pseries(void)
1726 {
1727         pr_info("Using radix MMU under hypervisor\n");
1728
1729         pseries_lpar_register_process_table(__pa(process_tb),
1730                                                 0, PRTB_SIZE_SHIFT - 12);
1731 }
1732 #endif
1733
1734 #ifdef CONFIG_PPC_SMLPAR
1735 #define CMO_FREE_HINT_DEFAULT 1
1736 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
1737
1738 static int __init cmo_free_hint(char *str)
1739 {
1740         char *parm;
1741         parm = strstrip(str);
1742
1743         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
1744                 pr_info("%s: CMO free page hinting is not active.\n", __func__);
1745                 cmo_free_hint_flag = 0;
1746                 return 1;
1747         }
1748
1749         cmo_free_hint_flag = 1;
1750         pr_info("%s: CMO free page hinting is active.\n", __func__);
1751
1752         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
1753                 return 1;
1754
1755         return 0;
1756 }
1757
1758 __setup("cmo_free_hint=", cmo_free_hint);
1759
1760 static void pSeries_set_page_state(struct page *page, int order,
1761                                    unsigned long state)
1762 {
1763         int i, j;
1764         unsigned long cmo_page_sz, addr;
1765
1766         cmo_page_sz = cmo_get_page_size();
1767         addr = __pa((unsigned long)page_address(page));
1768
1769         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
1770                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
1771                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
1772         }
1773 }
1774
1775 void arch_free_page(struct page *page, int order)
1776 {
1777         if (radix_enabled())
1778                 return;
1779         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
1780                 return;
1781
1782         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
1783 }
1784 EXPORT_SYMBOL(arch_free_page);
1785
1786 #endif /* CONFIG_PPC_SMLPAR */
1787 #endif /* CONFIG_PPC_BOOK3S_64 */
1788
1789 #ifdef CONFIG_TRACEPOINTS
1790 #ifdef CONFIG_JUMP_LABEL
1791 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
1792
1793 int hcall_tracepoint_regfunc(void)
1794 {
1795         static_key_slow_inc(&hcall_tracepoint_key);
1796         return 0;
1797 }
1798
1799 void hcall_tracepoint_unregfunc(void)
1800 {
1801         static_key_slow_dec(&hcall_tracepoint_key);
1802 }
1803 #else
1804 /*
1805  * We optimise our hcall path by placing hcall_tracepoint_refcount
1806  * directly in the TOC so we can check if the hcall tracepoints are
1807  * enabled via a single load.
1808  */
1809
1810 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
1811 extern long hcall_tracepoint_refcount;
1812
1813 int hcall_tracepoint_regfunc(void)
1814 {
1815         hcall_tracepoint_refcount++;
1816         return 0;
1817 }
1818
1819 void hcall_tracepoint_unregfunc(void)
1820 {
1821         hcall_tracepoint_refcount--;
1822 }
1823 #endif
1824
1825 /*
1826  * Keep track of hcall tracing depth and prevent recursion. Warn if any is
1827  * detected because it may indicate a problem. This will not catch all
1828  * problems with tracing code making hcalls, because the tracing might have
1829  * been invoked from a non-hcall, so the first hcall could recurse into it
1830  * without warning here, but this better than nothing.
1831  *
1832  * Hcalls with specific problems being traced should use the _notrace
1833  * plpar_hcall variants.
1834  */
1835 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
1836
1837
1838 notrace void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
1839 {
1840         unsigned long flags;
1841         unsigned int *depth;
1842
1843         local_irq_save(flags);
1844
1845         depth = this_cpu_ptr(&hcall_trace_depth);
1846
1847         if (WARN_ON_ONCE(*depth))
1848                 goto out;
1849
1850         (*depth)++;
1851         preempt_disable();
1852         trace_hcall_entry(opcode, args);
1853         (*depth)--;
1854
1855 out:
1856         local_irq_restore(flags);
1857 }
1858
1859 notrace void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
1860 {
1861         unsigned long flags;
1862         unsigned int *depth;
1863
1864         local_irq_save(flags);
1865
1866         depth = this_cpu_ptr(&hcall_trace_depth);
1867
1868         if (*depth) /* Don't warn again on the way out */
1869                 goto out;
1870
1871         (*depth)++;
1872         trace_hcall_exit(opcode, retval, retbuf);
1873         preempt_enable();
1874         (*depth)--;
1875
1876 out:
1877         local_irq_restore(flags);
1878 }
1879 #endif
1880
1881 /**
1882  * h_get_mpp
1883  * H_GET_MPP hcall returns info in 7 parms
1884  */
1885 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
1886 {
1887         int rc;
1888         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1889
1890         rc = plpar_hcall9(H_GET_MPP, retbuf);
1891
1892         mpp_data->entitled_mem = retbuf[0];
1893         mpp_data->mapped_mem = retbuf[1];
1894
1895         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
1896         mpp_data->pool_num = retbuf[2] & 0xffff;
1897
1898         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
1899         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
1900         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
1901
1902         mpp_data->pool_size = retbuf[4];
1903         mpp_data->loan_request = retbuf[5];
1904         mpp_data->backing_mem = retbuf[6];
1905
1906         return rc;
1907 }
1908 EXPORT_SYMBOL(h_get_mpp);
1909
1910 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
1911 {
1912         int rc;
1913         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
1914
1915         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
1916
1917         mpp_x_data->coalesced_bytes = retbuf[0];
1918         mpp_x_data->pool_coalesced_bytes = retbuf[1];
1919         mpp_x_data->pool_purr_cycles = retbuf[2];
1920         mpp_x_data->pool_spurr_cycles = retbuf[3];
1921
1922         return rc;
1923 }
1924
1925 #ifdef CONFIG_PPC_64S_HASH_MMU
1926 static unsigned long __init vsid_unscramble(unsigned long vsid, int ssize)
1927 {
1928         unsigned long protovsid;
1929         unsigned long va_bits = VA_BITS;
1930         unsigned long modinv, vsid_modulus;
1931         unsigned long max_mod_inv, tmp_modinv;
1932
1933         if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
1934                 va_bits = 65;
1935
1936         if (ssize == MMU_SEGSIZE_256M) {
1937                 modinv = VSID_MULINV_256M;
1938                 vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
1939         } else {
1940                 modinv = VSID_MULINV_1T;
1941                 vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
1942         }
1943
1944         /*
1945          * vsid outside our range.
1946          */
1947         if (vsid >= vsid_modulus)
1948                 return 0;
1949
1950         /*
1951          * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1952          * and vsid = (protovsid * x) % vsid_modulus, then we say:
1953          *   protovsid = (vsid * modinv) % vsid_modulus
1954          */
1955
1956         /* Check if (vsid * modinv) overflow (63 bits) */
1957         max_mod_inv = 0x7fffffffffffffffull / vsid;
1958         if (modinv < max_mod_inv)
1959                 return (vsid * modinv) % vsid_modulus;
1960
1961         tmp_modinv = modinv/max_mod_inv;
1962         modinv %= max_mod_inv;
1963
1964         protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1965         protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1966
1967         return protovsid;
1968 }
1969
1970 static int __init reserve_vrma_context_id(void)
1971 {
1972         unsigned long protovsid;
1973
1974         /*
1975          * Reserve context ids which map to reserved virtual addresses. For now
1976          * we only reserve the context id which maps to the VRMA VSID. We ignore
1977          * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1978          * enable adjunct support via the "ibm,client-architecture-support"
1979          * interface.
1980          */
1981         protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1982         hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1983         return 0;
1984 }
1985 machine_device_initcall(pseries, reserve_vrma_context_id);
1986 #endif
1987
1988 #ifdef CONFIG_DEBUG_FS
1989 /* debugfs file interface for vpa data */
1990 static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1991                               loff_t *pos)
1992 {
1993         int cpu = (long)filp->private_data;
1994         struct lppaca *lppaca = &lppaca_of(cpu);
1995
1996         return simple_read_from_buffer(buf, len, pos, lppaca,
1997                                 sizeof(struct lppaca));
1998 }
1999
2000 static const struct file_operations vpa_fops = {
2001         .open           = simple_open,
2002         .read           = vpa_file_read,
2003         .llseek         = default_llseek,
2004 };
2005
2006 static int __init vpa_debugfs_init(void)
2007 {
2008         char name[16];
2009         long i;
2010         struct dentry *vpa_dir;
2011
2012         if (!firmware_has_feature(FW_FEATURE_SPLPAR))
2013                 return 0;
2014
2015         vpa_dir = debugfs_create_dir("vpa", arch_debugfs_dir);
2016
2017         /* set up the per-cpu vpa file*/
2018         for_each_possible_cpu(i) {
2019                 sprintf(name, "cpu-%ld", i);
2020                 debugfs_create_file(name, 0400, vpa_dir, (void *)i, &vpa_fops);
2021         }
2022
2023         return 0;
2024 }
2025 machine_arch_initcall(pseries, vpa_debugfs_init);
2026 #endif /* CONFIG_DEBUG_FS */